Volume 7, Issue 4 p. 579-591
Original Research
Open Access

Wildfire severity and vegetation recovery drive post-fire evapotranspiration in a southwestern pine-oak forest, Arizona, USA

Helen M. Poulos

Corresponding Author

Helen M. Poulos

College of the Environment, Wesleyan University, Middletown, Connecticut, USA


Helen M. Poulos, College of the Environment, Wesleyan University, Middletown, CT 06459. Tel: (203)980-9773; E-mail: [email protected]

Search for more papers by this author
Andrew M. Barton

Andrew M. Barton

University of Maine at Farmington, Farmington, Maine, USA

Search for more papers by this author
George W. Koch

George W. Koch

Department of Biological Sciences and the Center for Ecosystem Science and Society, Northern Arizona University, Flagstaff, Arizona, USA

Search for more papers by this author
Thomas E. Kolb

Thomas E. Kolb

School of Forestry, Northern Arizona University, Flagstaff, Arizona, USA

Search for more papers by this author
Andrea E. Thode

Andrea E. Thode

School of Forestry, Northern Arizona University, Flagstaff, Arizona, USA

Search for more papers by this author
First published: 08 May 2021
Citations: 9


Post-fire stand water balance is a critical factor influencing tree regeneration and survival, which are often modulated by fire severity. We examined influences of the post-fire vegetation matrix and fire severity on diurnal, seasonal, and multi-year variation in evapotranspiration (ET) by analyzing the relationship between post-fire vegetation and ECOsystem Spaceborne Thermal Radiometer Experiment on the International Space Station (ECOSTRESS) ET data using multivariate and linear mixed effects modeling. Unlike many high-severity fire sites where ET drops after burning, post-fire ET was high at shrubland sites that burned at high fire severity in southern Arizona, USA. In this study, post-fire ET was driven by plant species composition and tree canopy cover. ET was significantly higher in the morning and midday in densely vegetated post-fire shrublands than pine-dominated forests that remained 5–7 years after wildfire. Our results demonstrate that plant functional traits such as resprouting and desiccation tolerance drive post-fire ET patterns, and they are likely to continue to play critical roles in shaping post-fire plant communities and forest water cycling under future environmental change.


Fire is a pervasive ecological disturbance that has spurred the evolution of a wide range of fire survival and post-fire recovery strategies in land plants (Keeley et al., 2011; Pausas et al., 2004; Pausas et al., 2006). While plants growing in fiery landscapes are ‘fire-adapted’, and although some display multiple fire-adapted traits, no species is adapted to all fire regimes (Pausas et al., 2017; Poulos et al., 2018; Schwilk, 2015; Schwilk & Ackerly, 2001). The recent anthropogenic intensification of wildfires in the western US (Dennison et al., 2014; Harvey, 2016; Westerling, 2016) is a novel biological filter that favors species capable of regenerating after high-severity fire over those adapted to more frequent, lower-severity fire regimes. This shift has resulted in fire-driven transitions from forests to shrublands at many sites across the western US (Coop et al., 2020; Guiterman et al., 2018; Haffey, 2014).

Historic lower-severity fire regimes maintained a matrix of fire-resistant tree species capable of regenerating only by seed (obligate seeders) intermixed with species also capable of regenerating by resprouting after fire (facultative resprouters) (Kaib et al., 1996; Kaib et al., 1998). However, the warmer, drier and more intense fires of the contemporary era have resulted in extensive mortality of thick-barked, obligate seeder conifers and a lack of post-fire seedling regeneration, especially at high fire severity sites lacking nearby seed trees (Haffey, 2014; Haffey et al., 2018; Stevens-Rumann & Morgan, 2019). In contrast, shrubs capable of vegetative regeneration by resprouting have continued to thrive in the wake of severe wildfire (Barton & Poulos, 2018; Guiterman et al., 2018; O’Connor et al., 2014).

Fire severity modulates ecosystem evapotranspiration (ET) via its effects on vegetation, microclimate, and water infiltration and run-off (Bosch & Hewlett, 1982; Dore et al., 2010; Nolan et al., 2014; Poon & Kinoshita, 2018; Wine & Cadol, 2016). Low-severity fires kill little aboveground vegetation, whereas high-severity fires kill nearly all aboveground vegetation, thereby reducing transpirational leaf surface area and increasing bare ground cover, soil evaporation, and surface runoff immediately after burning. High-severity fires can reduce ET and shift the balance of evaporation and transpiration for years to decades as vegetation recovers from burning (Ma et al., 2020; Montes-Helu et al., 2009; Poon & Kinoshita, 2018). Fire severity also influences microsite characteristics such as solar radiation, temperature and vapor pressure deficit, with concomitant impacts on site evaporation (E) (Aguilar et al., 2010; El Maayar & Chen, 2006). Typical ecosystem responses to high severity fire include increased E and decreased T, ET, and carbon uptake, as well as increased water surface run-off and stream flow (Dore et al., 2010; Nolan et al., 2014; Poon & Kinoshita, 2018; Wine & Cadol, 2016).

Post-fire species composition may influence ET because plant species differ in physiological characteristics and water use. In the western USA, for example, obligate seeding tree species such as pines strongly reduce stomatal conductance during drought to reduce T and minimize change in internal leaf water status (‘isohydry’), whereas stomatal conductance of oaks and other shrubs is less sensitive to drought which allows high T across a wide range of moisture conditions (‘anisohydry’) (Cavender-Bares & Bazzaz, 2000; Choat et al., 2012; Cooper et al., 2018a; Meinzer et al., 2009; Meinzer et al., 2014; Poulos et al., 2020). Thus, post-fire shrublands dominated by anisohydric oaks and shrubs may have higher T and possibly ET than forests containing a mixture of isohydric pines and anisohydric shrubs.

The impacts of burning on ecosystem ET have been examined only at severely burned sites with little or gradual post-fire vegetation recovery where reduced T and increased soil E dominate the post-burning response. Fire impacts on ET may operate differently at sites dominated by shrublands that have regenerated via resprouting in response to high-severity burning. We hypothesize that these conditions, acting through rapidly restored leaf area of anisohydric plant species, should result in higher T and ET and lower E in post-fire shrublands compared to forests. We examine this hypothesis using a new remote sensing approach for measurement of ET and its components – the ECOsystem Spaceborne Thermal Radiometer Experiment on the International Space Station (ECOSTRESS) (https://ecostress.jpl.nasa.gov/downloads/atbd/ECOSTRESS_L3_ET_PT-JPL_ATBD_20180509.pdf).

The ECOSTRESS mission offers improvements in remotely sensed measurement of ET and plant water stress. Remotely sensed imagery is being increasingly employed for examining ET patterns across landscapes (Ha et al., 2015; Hu et al., 2015; Jia et al., 2012; Long et al., 2014). Data products from these measurements have been developed from a variety of satellite sensors (Hu et al., 2015; Mu et al., 2005; Vinukollu et al., 2011), but these are currently available for only moderate spatial or infrequent temporal resolutions. ECOSTRESS uses a new, higher spatiotemporal resolution thermal infrared (between 8 and 12.5 μm), multispectral scanner on the International Space Station that provides high spatial resolution (70-m) data across the diurnal cycle because of its inclined, precessing orbit (Fisher et al., 2020; Hulley et al., 2017). Remotely sensed ET data acquired throughout the day offer the opportunity to evaluate the diurnal dynamics of stomatal closure in response to plant water stress imposed by seasonal drought and wildfire. A series of ecophysiological scalar functions based on atmospheric vapor pressure deficit, relative humidity and vegetation indices, including normalized difference and soil adjusted vegetation indices, are used to calculate actual ET and partition ET into canopy T and soil E, thereby facilitating investigation of complex controls over ET (Fisher et al., 2020). Here, we present the first ECOSTRESS characterization of post-fire ET with the specific objective of evaluating the interacting influences of post-fire plant species composition, stand structure, and fire severity on ET of a complex mountain landscape.

Materials and Methods

Study area

We conducted this study in the Chiricahua Mountains (Fig. S1) in southeastern Arizona (31°52’N, 109°15’W), which form part of the Sky Islands, a northern extension of the Sierra Madre Occidental (DeBano et al., 1995). The mountains extend southeast to northwest for about 80 km and rise from about 1100 to 3000 m altitude (a.s.l.). The terrain is rugged and highly dissected, rising from broad flat basins to rocky uplands, separated by steep-walled canyons. Soils are shallow, mostly derived from volcanic rhyolites and monzonites deposited in the early- to mid-Miocene, although pre-Tertiary rock is prominent at lower elevations (Drewes & Williams, 1984 [1973]).

The climate is semiarid, with two wet seasons, one between July and September, when more than 50% of total precipitation falls, and the second between December and March. A pronounced dry season usually occurs between the final winter storms in March or April and the onset of the North American Monsoon System in early July (Adams & Comrie, 1997). At the Chiricahua National Monument Visitor Center (1650 m a.s.l.), mean monthly minimum and maximum temperatures range from −1.2°C to 13.4°C in January to 15.5°C and 31.8°C in July; mean annual precipitation is 483 mm.

We conducted this research in Cave Creek Canyon (CCC, 1600–2165 m a.s.l.) in the Coronado National Forest on the east side of the Chiricahua Mountains and in Chiricahua National Monument (CNM) on the west side (Fig. S1). Fire histories have revealed a complex fire regime across vegetation types before Euro-American settlement (<1880s), with mainly frequent surface fires in forests (Kaib et al., 1996; Kaib et al., 1998; Swetnam et al., 2001) and more mixed fire regimes, with longer fire intervals on uplands and drier sites (Kaib et al., 1996, Morino et al., 2000, Minor et al., 2017). As elsewhere in the American Southwest, livestock grazing and then fire suppression largely excluded fire starting in the 1880s up to late in the 20th century, followed by a contemporary period of more frequent, much larger, and more intense wildfires. During the extremely dry year of 2011 (Williams et al., 2014), the Horseshoe Two Fire burned >90 000-ha, about 75% of the Chiricahua Mountains, with 12% burning at high- and 30% at moderate-severity (Arechederra-Romero, 2012). Sites that burned at moderate to high severity experienced a decrease in pines and increase in oaks and other taxa capable of resprouting after wildfire (Barton & Poulos, 2018).

Field vegetation data

We sampled vegetation in 226 forest monitoring transects spanning the fire severity gradient (63 in CCC and 163 in CNM) 5–7 years after the 2011 Horseshoe Two Fire (Fig. S1). In each 5 × 25-m (0.125-ha) belt transect, we recorded the percentage cover by woody plant species in successive 5 × 25-m blocks. Percent cover by species was then calculated as the mean % cover by species over the entire 25-m transect.

We investigated woody plant species patterns using cluster analysis and ordination. First, we performed hierarchical cluster analysis on the cover data using Ward’s method in R (R Development Core Team, 2020) to identify the dominant post-fire plant cover types in our dataset. The optimal number of vegetation types (or clusters) was determined using the NbClust() function in the NbClust package (Charrad et al., 2014). We then employed indicator species analysis via the indicspecies package (Caceres et al., 2016) to ascribe names to each vegetation type, based on the fidelity of appearance of species in each vegetation type cluster. Finally, we used the vegan package (Oksanen et al., 2007) to perform non-metric multidimensional scaling (NMDS) on the transect cover data to (1) explore how post-fire species composition varied by vegetation type, (2) evaluate how post-fire woody plant species composition was related to fire severity, and (3) generate univariate estimates of species composition for each plot. We employed NMDS as a multivariate data reduction technique to evaluate rank-order dissimilarities in woody plant species composition among vegetation transects. NMDS is preferable for analyzing ecological datasets over the more widely used principal components analysis ordination technique because it minimizes distortions in the final ordination solution by optimizing and preserving the distances of individuals among observations (vegetation transects in woody plant cover species space, in this case) (Kenkel & Orlóci, 1986). For this analysis, we examined both the degree to which the first two NMDS axes explained variation in community composition, the loadings of individual species along axes to reveal community variation in species composition among plots, and the correlation between fire severity and the first two NMDS axes.

Remotely sensed data

We evaluated the influence of fire severity on post-fire ET using raster delta normalized burn ratio (dNBR) fire severity data (Eidenshink et al., 2007) for the 2011 Horseshoe Two Fire for each of the 226 vegetation transects. dNBR is a Landsat ETM+ derived product that measures absolute change in fire severity that is correlated to the pre-change image. The normalized burn ratio is calculated from ETM+ bands 4 and 7 as (ETM4 − ETM7)/(ETM4 + ETM7), where ETM4 represents the near-infrared spectral range (0.76–0.90 μm) and ETM7 represents the shortwave infrared spectral range (2.08–2.35 μm). Differenced NBR images (post-fire NBR subtracted from pre-fire NBR) are referred to as dNBR images. Pre-fire Landsat ETM+ images are taken from the month before the fire and post-fire images are taken 6 months post-fire for dNBR calculation. We chose to use dNBR rather than other relativized remotely sensed fire severity products because absolute change is a better measure than relative change for our study as it relates more directly to how much vegetation cover change occurred in response to the fire, which was our predicted driver of ET. Fire severity data were acquired from the Monitoring Trends in Burn Severity data distribution site (https://www.mtbs.gov/), and dNBR values for transect locations were extracted using the point sampling tool in QGIS (QGIS Development Team, 2020).

We obtained ET data for each vegetation transect from the ECOSTRESS PT-JPL ET product for cloud-free images between February 2018–2020. The ECOSTRESS PT-JPL ET product is derived from 70 m ECOSTRESS imagery with inputs of atmospheric and surface data from MODIS and vegetation cover properties from Landsat (Fisher, 2018). This estimate of ET (as the latent heat flux) includes the ET components of canopy transpiration, soil evaporation and interception evaporation using the Priestley-Taylor (PT) JPL retrieval algorithm, as described in Fisher (2018). We extracted ET data for each transect location using the AppEEARS LP-DAAC application (AppEEARS, 2019). Cloud masks were evaluated prior to further analysis by interpreting the bit mask reports for each image. Only cloud-free images were included in all further analyses. We used the ECOSTRESS-based ET estimate rather than alternatives, such as a purely MODIS-based measure, because its 70 m resolution (vs. 250 m for MODIS) is crucial for measurements in fine-grained environments, such as the highly dissected terrain of the Chiricahua Mountains.

Lastly, we explored how dNBR interacted with post-fire canopy cover and vegetation greenness, or ‘leafiness’ to modulate ET by extracting National Land Cover Dataset (NLCD) tree canopy cover data and Landsat 8 annual composite enhanced vegetation index (EVI) data for each vegetation transect location. We obtained 30 m resolution NLCD percent tree canopy cover data for 2016 (the most recent release of this remotely sensed product) from the Multi-Resolution Land Characteristics Consortium (https://www.mrlc.gov/data). The 2016 NLCD tree canopy cover dataset ranges from 0% to 100%, and the product was generated using Landsat 8 scenes and US Forest Service Forest Inventory and Analysis data via random forest modeling (Yang et al., 2018). Landsat 8 2018 EVI composite data were obtained from Google Earth Engine (https://explorer.earthengine.google.com), which is calculated as 2.5 × ((Band 5 − Band 4)/(Band 5 + 6 × Band 4 − 7.5 × Band 2 + 1)). The EVI is a widely used remotely sensed proxy for vegetation greenness that is responsive to canopy structural variations, including leaf area index, canopy type, plant physiognomy and canopy architecture (Huete et al., 2002), all of which are likely to vary by fire severity in the post-fire landscape. We used the 2018 annual EVI composite for our modeling procedure to coincide with the field vegetation sampling interval.

Mixed effects models

We performed a series of linear mixed effects analyses (via the lme4 package; Bates et al., 2013) to evaluate the relationship of ECOSTRESS ET with fire severity, post-fire woody plant species composition (NMDS scores), percent canopy cover, and EVI. Separate models were fit to evaluate these influences over the diurnal cycle (morning 8–11 am, midday 11 am–2 pm, afternoon 3–5 pm and evening 5–8 pm), in different seasons (winter, spring, summer, fall; using the equinox to categorize observations by season), among vegetation types, and over the entire two-year time-series. Daily total, canopy, and soil ET were used as the response variables for the seasonal, vegetation type and full time-series analyses, while we specified instantaneous ET as the response variable for all analyses of diurnal ET variation. We used linear mixed effects models to account for the covariance structures in the data from the repeated measures of ECOSTRESS ET at each sample plot. Random effects were designated for the intercept and the slope of the ECOSTRESS observations, which were nested within sample transects in all models. Fixed effects included time of day, season and vegetation type. We inspected the residuals of each model for deviations from homoscedasticity and transformed data as necessary; all final models contained residuals without obvious deviations from normality. Final effects plots for each model were generated using the effects package (Fox & Weisberg, 2018). Post-hoc pairwise comparisons and interaction terms were analyzed for all models using the jtools package (Long & Long, 2017).

Results and Discussion

This study is the first to use ECOSTRESS to examine the impacts of wildfire on ecosystem water balance. Our goal was to scale up from the post-fire physiological responses of individual plants (e.g. Poulos et al., 2020) to the landscape through the integration of field vegetation data and remotely sensed estimates of fire severity and ecosystem water balance. We acknowledge that ECOSTRESS imagery is a proxy for more direct ET measurements, such as with field eddy covariance or sapflow. Future field validation of ECOSTRESS performance in post-wildfire landscapes will be required to enhance our understanding of the utility of the ECOSTRESS product. However, the frequent and large spatial coverage of the ECOSTRESS sensor provides an important opportunity for evaluating how landscape-scale patterns of post-fire vegetation influence ET. The results of this study highlight the complexity of how fire severity and the resulting post-fire vegetation matrix interact to influence ET. Our study is also the first to demonstrate that post-fire ET can be high on high-severity fire sites with prolific post-fire regeneration by anisohydric, sprouting shrubs, results that contrast with prior research reporting low ET after high severity forest fire (e.g. Dore et al., 2010; Ma et al., 2020; Poon & Kinoshita, 2018).

Influences of the post-fire vegetation matrix on daily ET

We identified three dominant post-fire vegetation types from the field-derived plant species cover data: shrubland (n = 71), pine-oak forest (n = 89) and piñon scrub (n = 66) (Figs. 1A and 2; Table S3). Shrublands were dominated by Qhypoleucoides and Ceanothus fendlerii, while major pine-oak forest species included thick-barked pines (Pinus engelmannii, and Pleiophylla) and Quercus arizonica (Figs. 1A and 2). Piñon scrub was comprised mainly of dry site specialist species, especially P. discolor, Qtoumeyii and Arctostaphylos pungens. NLCD tree canopy cover and EVI differed significantly among the three post-fire vegetation types (P < 0.05, Table S1; Fig. S2). Post-fire shrublands had the highest canopy cover and EVI (35.9% ± 0.9 se and 0.45 ± 0.09 se, respectively), pine-oak forest was intermediate (25.6% ± 0.6 se and 0.37 ± 0.11 se, respectively.) while piñon scrub displayed significantly lower canopy cover and EVI than the other two vegetation types (18.3% ± 0.2 se and 0.25 ± 0.06 se, respectively).

Details are in the caption following the image
Influences of post-fire vegetation on daily evapotranspiration (ETday) in Chiricahua National Monument, Arizona. Panels depict: (A) The non-metric multidimensional scaling of 226 post-fire vegetation transects (gray dots) in species space with vegetation types plotted with 95th percentile confidence ellipses around the centroid of each vegetation type. Fire severity (dNBR) is significantly correlated with NMDS axes 1 and 2 (P < 0.001) as shown by the dNBR vector; pine-oak forest experienced lower fire severity, while shrublands were subject to higher-severity fire; (B) significant influences of NMDS1 and fire severity on daily ET (ETday); (C) significant NMDS1 axis effects on canopy transpiration (Tinst) (but the impact of fire severity was not significant); and (D) NMDS1 axis effects and fire severity influences soil evaporation (Einst) (P < 0.05). Woody plant species acronyms are described in Table S4. Species composition is significantly correlated with NMDS1 in panels (B–D). dNBR, delta normalized burn ratio; NMDS, non-metric multidimensional scaling.
Details are in the caption following the image
Photos of the three dominant post-fire vegetation types 5–7 years after the 2011 Horseshoe Two Fire. Pine-oak forests and piñon scrub maintain a mixture of pines and oaks after wildfire, while post-fire shrublands are characterized by standing dead pines that were killed by the fire and post-fire recovery by resprouter tree species.

Piñon scrub displayed significantly lower ET and canopy T than did pine-oak forest, while shrublands displayed high variation in ET and canopy T, and therefore neither ET nor canopy T differed significantly between shrublands and the two other vegetation types (Tables S3 and S4). In contrast, soil E differed significantly among all vegetation types as follows: piñon scrub > pine-oak forest > shrublands (P < 0.05); a pattern that is likely also associated with the negative relationship between tree canopy cover and soil E (Table S1; Fig. S2).

Vegetation composition effects on daily ET

NMDS axis 1 represents a gradient from shrubland (Cfendlerii and Q. hypoleucoides) to pine-oak forest (Pengelmannii, P. leiophylla, and Qhypoleucoides) to piñon scrub (P. discolor, Apungens, and Q. toumeyi) (Figs. 1A and 2; Table S5). Figure 1A shows that fire severity increased from high to low along NMDS Axis 1, and from low to high NMDS Axis 2. Mean dNBR was 330 ± 19.7 se in piñon scrub, 390 ± 18.9 in pine-oak forest, and 602 ± 32.3 in shrublands (dNBR range was 9–933 across all plots), meaning that shrubland plots were by far the most common vegetation type in higher-severity fire plots.

Post-fire ET was significantly related to plant species composition as expressed by the NMDS axis 1 scores. Canopy T decreased from lower to higher NMDS scores, that is, from mainly shrubland to pine-oak forest to piñon scrub vegetation types (Fig. 1C). The effect of species composition on soil E was amplified with increasing fire severity, such that there was little effect of species composition on soil E after lower-severity fire but pronounced impacts at higher severity (Fig. 1D). The pattern for ET was the opposite: the effect of species composition was higher for lower severity fire than for sites subject to higher severity fire (Fig. 1B). These results demonstrate the importance of woody plant species composition on post-fire ET generally and the higher post-fire T and ET of shrublands and pine-oak forests with high canopy cover relative to piñon scrub. These results also highlight that soil evaporation is an important component of ET in piñon scrub sites that typically have large areas of exposed bare soil and rock compared to the more densely vegetated pine-oak and shrubland cover types (Fig. 1D; Tables S1 and S4).

Fire severity as a driver of daily ET

Fire severity was surprisingly not a significant driver of post-fire daily ET over the entire two-year time-series (Fig. 3A; Table S6). This result likely stemmed from the complex interactions between fire severity and post-fire vegetation composition and cover. The relationship between canopy transpiration and dNBR was positive, but not significant (P = 0.39). However, soil evaporation (E) was a significant driver of post-fire ET in this semi-arid landscape (Fig. 3; Table S6).

Details are in the caption following the image
Linear mixed effects model results of fire severity (dNBR) versus (A) daily evapotranspiration (ETday), (B) canopy transpiration (T) and (C) soil evaporation (E) in CNM. Rug plots on the x-axis of each panel display the sample depth of the dNBR observations over the fire severity gradient. Daily soil evaporation declined significantly with increasing fire severity. Full model results are shown in Table S6. dNBR, delta normalized burn ratio.

First, the lack of a significant relationship of daily ET and T with fire severity can potentially be explained by the way in which wildfires alter forest stand structure in systems where high severity fire promotes post-fire resprouting. High ET occurs in densely vegetated areas, regardless of forest cover type, in accordance with the well-established leaf area index-ET relationship (Kristensen 1974). The significant positive relationship between tree canopy cover and canopy T and EVI in this study corroborates this pattern (Table S1; Fig. S2).

Our results differ from earlier studies of the impacts of high-severity burning on forest ET in the western U.S. In these earlier studies, high fire severity resulted in low plant canopy cover and decreased ET, primarily due to decreased vegetation transpiration (Blount et al., 2020, Dore et al., 2010, Ha et al., 2015, Ma et al., 2020, Montes-Helu et al., 2009, Poon & Kinoshita, 2018). Vegetation recovery at these high fire severity sites was slow, with persistent low cover and high soil exposure resulting in high levels of soil E but low canopy T and overall ET, sometimes lasting for 15 years or more.

In our study, plots that did not burn or burned at low severity maintained a largely undisturbed vegetation matrix, and most of these sites retained large trees and canopy cover after the fire. As in prior studies, these stands exhibited high post-fire ET (Fig. 3A), but unlike other studies on post-fire ET patterns at high severity sites, post-fire shrublands that regenerated in the wake of high-severity fire had high post-fire vegetation cover and, as a result, also displayed high ET. These divergent mechanisms in sites experiencing low versus high fire severity both led to high post-fire ET, explaining the lack of significant differences in ET and T across the fire severity gradient. In fact, when the data were analyzed by season, ET increased significantly from low to high severity sites for summer, the main growing season, and summer ET was also significantly higher in shrublands relative to the two other vegetation types (P < 0.05) (Fig. 4; Table S6).

Details are in the caption following the image
Seasonal mixed effects model results of (A) seasonal differences in daily evapotranspiration (ET) by fire severity (dNBR), and (B) summer ET by vegetation type. Rug plots on the x-axis of each panel display the sample depth of the dNBR observations over the fire severity gradient. Both models are significant at the P < 0.05-level. Growing season (summer) daily ET is significantly higher than other times of year, and summer ET is significantly higher in shrublands relative to the two other vegetation types at moderate to high fire severities. dNBR, delta normalized burn ratio.

Unlike T and ET, soil Eday significantly declined across the fire severity gradient (Fig. 3C; Table S2 and S3). The arguments given previously for why ET and T did not decline very likely apply here as well. Prolific resprouting after higher severity fire rapidly provided vegetative cover, reducing exposed soil, which reduced soil E. As shown in the previous section, most of these plots were dominated by dense shrublands. Lower severity sites included many piñon scrub plots, which contained large areas of bare soil interspersed between short-statured Q. toumeyi resprouts where the ET signal was dominated by soil E (P < 0.05) (Fig. S3).

Diurnal variation in instantaneous ET

The precessing orbit of ECOSTRESS provided the opportunity to elucidate the effects of fire severity on diurnal instantaneous ET variation for the first time using remotely sensed imagery. Instantaneous post-fire ETinst varied significantly throughout the day (Fig. 5A; Table S7). However, morning ETinst was significantly modulated by fire severity only in the morning hours of the day from 8–11 am (Fig. 5A) (P < 0.001). As in the full daily ET dataset analysis, soil Einst was likely the driver of variation in morning post-fire ETinst over the fire severity gradient, and soil Einst decreased significantly with increasing fire severity, and from morning to midday as soils dried over the course of the day (Fig. 5C; Fig. S3). Likewise, vegetation type was an important influence on diurnal variation in ETinst, as demonstrated by a significant increase from morning to midday ETinst in shrublands (t = 3.04; P < 0.001) and the significantly higher midday ETinst in shrublands compared to the other two vegetation types (P < 0.001) (Tables S5 and S7).

Details are in the caption following the image
Effects plots of (A) diurnal variation in instantaneous evapotranspiration (ETinst), which differed across the day, (B) morning canopy transpiration (Tinst) (C) morning soil evaporation (Einst) were positively correlated, but not significant (P = 0.19), and (D) mean (±se) ETinst for morning and midday by vegetation type. Rug plots on the x-axis of panels (A–C) display the sample depth of the dNBR observations over the fire severity gradient. Morning ET decreased with increasing fire severity, and midday ET displayed the opposite pattern. Morning soil E decreased with increasing fire severity. The asterisk (*) in panel (D) denotes both a significant increase in ETinst from morning to midday in shrublands and the significantly higher midday ETinst of shrublands compared to other vegetation types at P < 0.01. Full model results are shown in Table S7. dNBR, delta normalized burn ratio.

These contrasting patterns could be the result of the complex dynamics of post-fire vegetation coupled with differences across species in diurnal gas exchange patterns. Overall, the decreasing trend of soil Einst with increasing fire severity suggests that rapid vegetation recovery at higher fire severity in dense post-fire shrublands ameliorates soil E. Seasonal ETinst patterns were further modulated by vegetation type: shrublands maintained significantly higher summer ETinst than any other vegetation type as fire severity increased. We have shown previously that plots subject to lower fire severity tend to support a mix of anisohydric oaks and desiccation avoidant, isohydric pines (Barton & Poulos, 2018; Poulos et al., 2020). Because the pines do not generally survive higher severity fire and rely largely on regeneration from seed, oak resprouts dominate sites after severe fires in our study site. These compositional differences appear to play out in the scaling of gas exchange from individual plants to entire stands. In the morning, the stomata of all species are open and transpiration is uniformly high (i.e. we found no significant differences in morning ET among vegetation types, as shown in Fig. 5D). By midday, the isohydric pines have reduced stomatal conductance and transpiration, whereas oaks continue to transpire at high levels (Poulos et al., 2020). At the stand level, ECOSTRESS morning ETinst is higher in the largely intact vegetation of lower severity sites compared to the recovering vegetation of higher severity sites. By midday, however, the isohydric pines have apparently reduced T, resulting in relatively low stand level ET in low severity sites, whereas oak dominated high severity sites continue to exhibit high levels of Tinst and ETinst. Maintaining high levels of stomatal conductance, transpiration and photosynthesis for much of the day are a well-documented strategy during post-fire recovery (Moya et al., 2015; Nolan et al., 2015), and anisohydric resprouters generally display significantly higher daily stand-scale ET than obligate seeder pine species (Clemente et al., 2005; Cooper et al., 2018a, b).

Fire severity, post-fire vegetation, and stand-level ET

This research contributes to the growing body of work on the consequences of wildfire-driven vegetation type conversions in western North America. Our results support two related hypotheses. First, fire severity and post-fire plant community composition significantly control temporal variation in ET in Madrean pine-oak forests. Second, vigorous post-fire shrublands in severely burned sites in the Chiricahua Mountains exhibit high stand-level ET due to rapid recovery of leaf area via vegetative resprouting and the anisohydric behavior of oaks that dominate such scrublands. These results support our hypotheses about controls on post-fire ET in Madrean pine-oak ecosystems, and they represent an important addition to our previous understanding about post-fire ET, which has been largely shaped by studies in ecosystems with low levels of post-fire vegetation recovery and associated low ET that can remain depressed for decades (Dore et al., 2010; Ma et al., 2020; Montes-Helu et al., 2009; Poon & Kinoshita, 2018).

The mechanism likely underlying our results is the ability of shrub species, the dominant species group after high severity fire, to resprout vigorously from previously provisioned underground tissues, rapidly recover leaf area, and generate high transpiration compared to the mixed vegetation of sites subject to lower fire severity (Poulos et al., 2020). As in shrublands, piñon scrub sites also experienced a loss of pines and recovery by resprouter shrubs, but post-fire resprouting within this vegetation type was more limited likely due to the hot and dry conditions that prevail in these sites. Our results also raise the possibility that differences in the physiological capacity and response of species—in this case, anisohydric oaks versus isohydric pines—may have contributed to the stand-level differences in ET in our study. The extent to which such individual species differences generally scale up to ecosystem level water relations deserves further attention (see, for example, Sun et al., 2020).

An increase in wildfire frequency and intensity driven by climate change in the American Southwest may increase dominance by fire-resilient sprouting shrublands. Our results offer important implications about vegetation resilience to climate change indicating that resprouter shrublands may be a more resilient vegetation type under a current and future intensified fire regime in this region (Falk, 2013; Falk et al., 2019). Obligate seeder species can take decades-to-centuries to recover from fire due to the episodic nature of conditions favorable for seedling establishment and growth (Chambers et al., 2016; Korb et al., 2019). Increasing aridity and more intense fire regimes may further hinder such regeneration, even if seed sources are nearby (Kemp et al., 2019; Liang et al., 2017). Thus, plant traits such as resprouting, water use strategy, and desiccation tolerance, which drive the responses documented here, are likely to continue to play critical roles in shaping post-fire plant communities and forest water cycling under future environmental change (Gratani, 2014).


Forests in the desert Southwest comprise a critical water-provisioning ecosystem service in an arid region (Boyanova et al., 2017; Keith et al., 2017; O'Donnell et al., 2018). Most projections indicate that western North America will continue to experience increasing aridity and larger, more severe, and more frequent wildfires (Abatzoglou & Kolden, 2013; Adams, 2013; Ault et al., 2016; Cook et al., 2015; Singleton et al., 2019), which may threaten this water resource. The intensification of fire-climate interactions over the coming decades could lead to the persistence of wildfire-generated shrublands (Loehman et al., 2018; O’Connor et al., 2020) which have surprisingly high ET, according to our results. Consequences of high post-fire ET are likely to include high carbon sequestration (Biederman et al., 2017) and a reduction of water provisioning to humans via runoff and drainage, which is supplied by water not used in ET. Our results of high ET of post-fire shrublands suggest no increase in water delivery from Madrean pine-oak forests for human uses following severe wildfire, which could be a cause for concern in a future, fire-intensified world.

The results from this study also highlight the utility of ECOSTRESS ET imagery products for evaluating spatiotemporal variation in stand water use in arid, wildfire-affected landscapes. The frequent image acquisition (once or twice per week), ecologically relevant spatial resolution (70 m), and the precessing orbit of the sensor provide gridded ET data that can be used to understand the mechanisms of vegetation recovery after wildfire and water balance across vast landscapes. As aridity and wildfire activity increase in the coming decades, understanding post-fire ET as a metric of vegetation recovery and water cycling will become increasingly important for guiding forest management and restoration activities, especially in the wake of high-severity wildfire.


The authors greatly appreciate the American Museum of Natural History’s South-western Research Station, the Douglas District of Coronado National Forest, and Helen Fitting and Mike Holt at Chiricahua National Monument for their logistical support of the study. We are grateful to Rob Kabacoff at Wesleyan University for his statistical consulting on the linear mixed effects model analysis. Finally, we thank field assistants Michael Donnelly, Zak Edwards, Charlie Faires, Michael Freiburger, Wyatt McCurdy, Sonya Sternleib, Rosie Wilkin, Isaiah Wilson-McFarlane and Hunter Vannier for invaluable assistance with the vegetation resampling effort in Arizona. Funding for this research was provided by NASA (Award 80NSSC20K0077), Western National Parks Association (2017), the National Park Service (Task Agreement P17AC00940), the Joint Fire Sciences Program (Award L15AC00152), the University of Maine at Farmington, and Wesleyan University’s Robert Schumann Institute of the College of the Environment. The authors state no conflict of interest in submitting this paper for consideration for publication.

    Funding Information

    Funding for this research was provided by NASA (Award 80NSSC20K0077), Western National Parks Association (2017), the National Park Service (Task Agreement P17AC00940), the Joint Fire Sciences Program (Award L15AC00152), the University of Maine at Farmington, and Wesleyan University’s Robert Schumann Institute of the College of the Environment.